Freethought & Rationalism Archive

S.C.Carlson · 09-06-2005, 03:57 PM

Quote:

Originally Posted by Ben C Smith

A follow-up.... From your description it appears that if both 020 and 002 (just my running example) lacked a particular word that happened to appear somewhat frequently in other parts of the synoptic traditions, the correlation between 020 and 002 would be improved, as it were. Is that correct?

Yes. Nulls are certainly a concern for the reason you state but there are standard techniques to deal with them. I do not recall whether Gentile used any of them. At any rate, it is not an intractable problem.

Quote:

Originally Posted by Ben C Smith

Also, it appears to me that this method really deals only with the broadest level of synoptic relations. For example, it does not appear capable of identifying any given pericope within 002 that may actually more closely resemble 200 than 020. Correct?

That's really an issue with the nature of the data that is fed into the method, rather than the method itself. Gentile used the statistics collected by Paul Hoffmann et al.'s Synoptic Concordance, but there's no reason why one cannot look at individual pericopes in more detail with his method--assuming, of course, that the sample size of the data doesn't become so small as to result in a major loss of statistical significance. So, it's more of an issue of the data, not the method.

Stephen

GentDave · 09-07-2005, 08:28 AM

Quote:

Originally Posted by Ben C Smith

Nice to meet you, Dave! Thanks for dropping in.

The problem is not yours; it is mine. I just know next to nothing about statistics (as a mathematical field). So when you start explaining the processes on your site in terms such as...:

...well, that pretty much goes right over my head.

I asked if the math could be dumbed down a bit, and the impression I am getting is that no, it is already as simple as it is going to get.

Let me ask one very specific question and see if the answer still eludes me. Suppose a certain Greek word is found 10 times in 020 material. Does your process try to predict, so to speak, that this same word should be found 10 times in 002 material too, and then, failing that, draw the conclusion that 020 and 002 material must not be very closely related (after, of course, 800 other words have been similarly tried)? Also, if the above is even slightly accurate as a description of your process, does it take into account that the 020 material might be more or less lengthy than the 002 material?

BTW, I have been drawn of late to the 3SH (in some form or other) for reasons completely unbased in statistics, word counts, or vocabulary comparisons. Which is why I asked about your site in the first place.

Thanks.

Ben.

Thanks.

On Poisson vs. Normal - A normal distribution is the standard bell curve that we hear about a lot. It can have positive and negative values, and non-integer values. A Poisson distribution can not have negative numbers, and can only have integer values. So, if you are talking about something like word counts (or individuals arriving at an ATM in textbooks), it is a good choice. Word counts are not negative or fractional. If you know a process should be described by a Poisson, and you know the average rate (of say arrivals at an ATM), you can give the probability of some other number of arrivals. Say if you normally get 5 arrivals in 5 minutes, you can give the probability of 10 arrivals.

Stephen did a nice job with most of your question. You ask -does it take into account that the 020 material might be more or less lengthy than the 002 material? And I can reaffirm the answer is yes, that is taken into account.

I think based on the study alone, without taking into account other factors, the 3SH looks best. Lately for reasons outside the study, I tend to think that Luke did not use Matthew originally, but later a lot of Matthian material made its way over.

GentDave · 09-07-2005, 08:44 AM

Quote:

Originally Posted by S.C.Carlson

Yes. Nulls are certainly a concern for the reason you state but there are standard techniques to deal with them. I do not recall whether Gentile used any of them. At any rate, it is not an intractable problem.

That's really an issue with the nature of the data that is fed into the method, rather than the method itself. Gentile used the statistics collected by Paul Hoffmann et al.'s Synoptic Concordance, but there's no reason why one cannot look at individual pericopes in more detail with his method--assuming, of course, that the sample size of the data doesn't become so small as to result in a major loss of statistical significance. So, it's more of an issue of the data, not the method.

Stephen

There is no special procedure for dealing with nulls.

To Stephen - They are treated as zeros, not nulls.

To the non-statistical - Nulls in statistics usually mean "missing data", and they can be a tricky issue, and require special care. In my study there are no missing values, but there are zero counts. The information is not missing, we know for a fact there are zero of that type of word in that category.

Yes, if two categories both lack a word that appears elsewhere, that will tend to draw them together and make them look more similar, but it's the relative lack of the word that is important. An example - suppose based on a control we expect 1 occurrence of the word "cat". If we examine the 2 categories and find zero cats in both, then that increases their similarity, but only a tiny bit. Now if we expected to find 20 cats, based on the control, and we found zero cats in both of the two categories, then that is a much stronger indication of similarity.

The technique could be applied to smaller samples of text, if the word counts were acquired, but the results would likely lack significance.

Ben C Smith · 09-07-2005, 10:59 AM

Quote:

Originally Posted by GentDave

On Poisson vs. Normal - A normal distribution is the standard bell curve that we hear about a lot. It can have positive and negative values, and non-integer values. A Poisson distribution can not have negative numbers, and can only have integer values. So, if you are talking about something like word counts (or individuals arriving at an ATM in textbooks), it is a good choice. Word counts are not negative or fractional. If you know a process should be described by a Poisson, and you know the average rate (of say arrivals at an ATM), you can give the probability of some other number of arrivals. Say if you normally get 5 arrivals in 5 minutes, you can give the probability of 10 arrivals.

Ah, much better.

Very helpful. You know, I could kind of be your guinea pig if you ever do decide to publish. Just give me a copy of the statistical parts, and if I can understand it anybody can.

Thanks.

Ben.

Peter Kirby · 09-07-2005, 11:07 AM

Why not publish in a journal on computational linguistics rather than a journal of theology or NT studies? The former will not care so much if you don't give a history of the Synoptic Problem, and it will give you greater latitude for describing the mathematics.

kind thoughts,
Peter Kirby

GentDave · 09-07-2005, 08:32 PM

Quote:

Originally Posted by Peter Kirby

Why not publish in a journal on computational linguistics rather than a journal of theology or NT studies? The former will not care so much if you don't give a history of the Synoptic Problem, and it will give you greater latitude for describing the mathematics.

kind thoughts,
Peter Kirby

Not a bad thought. Thanks for the suggestion.

Thread Tools	Search this Thread
Show Printable Version	Search this Thread: Advanced Search

09-07-2005, 11:07 AM	#15
Peter Kirby Veteran Member Join Date: Jul 2001 Location: the reliquary of Ockham's razor Posts: 4,035	Why not publish in a journal on computational linguistics rather than a journal of theology or NT studies? The former will not care so much if you don't give a history of the Synoptic Problem, and it will give you greater latitude for describing the mathematics. kind thoughts, Peter Kirby

Freethought & Rationalism Archive

The archives are read only.